RVC-Project/Retrieval-based-Voice-Conversion-WebUI

Error in epoch rounds, recounting from 1

Open

#1,954 创建于 2024年4月7日

在 GitHub 查看
 (0 评论) (0 反应) (0 负责人)Python (18,427 star) (2,849 fork)batch import
help wanted

描述

During the training process of my model, because I did not reserve enough space on the hard drive, the model stopped with an error "OSError:No space left on device". I cleaned it up and left enough space to start training again, but it should have trained for more than 400 rounds, but the epoch is counting from 1. Is this because the checkpoints are missing? Is there any negative impact on my training? I really need to solve this problem urgently, thanks for your help!

贡献者指南

技术栈
python
领域
machine learning
议题类型
research
难度面向新贡献者的预计实现难度，1 表示很小改动，5 表示专家级工作。
3
预计时间有经验贡献者完成调查、实现、测试并准备 pull request 的粗略时间范围。
1-3 hours
活动状态议题当前的可参与程度：新鲜、活跃、陈旧、阻塞或等待维护者输入。
fresh
清晰度议题是否清楚说明期望改动、验收标准和下一步。
unclear
前置要求
basic knowledge of training resumptionfamiliarity with checkpointing
新手友好度1-100 的估计分数，表示该议题对首次贡献者的友好程度。
40
研究方向
Investigate the training script's checkpoint loading logic, particularly the epoch counter. The user experienced a training interruption due to disk space, and after cleanup, training restarted from epoch 1. Check if the checkpoint file was corrupted or missing. Look at the training loop code to understand how epoch state is saved and restored. Provide guidance on how to resume training from the last saved epoch.